Database Reference
In-Depth Information
Algorithm 2.
Function Explore-Select-Rearrange(z, Q)
Z
res
← Null
Z′ ← Null
if Corr(z,Q) = indecisive then
for all child node z
child
of z do
Z′ ← Explore-Select-Rearrange(z
child
,Q)
if Z′ Null then
Z
res
.addChild (Z′)
end if
end forif Z
res
.NumberofChildren() > 1 then
Z
res
.BuildIntent()
Else
Z
res
←Zres.uniqueChild()
end if
else
if Corr(z,Q) = exact then
Z
res
← z
end if
end ifreturn Z
res
groups that share the same (or similar) vocabulary as theirs. In addition, they have to be familiar with
their group's linguistic labels before using them accurately. As a result, this option is not much more
convenient than using ad-hoc linguistic labels predetermined by a domain expert. Moreover, it only
transposes the problem of user-specific summaries maintenance to group-specific ones.
The second alternative, investigated in the following, consists in building only one SAINTETIQ sum-
mary hierarchy using an ad-hoc vocabulary, and querying it with user-specific linguistic labels. Since
the vocabulary of the user's query
Q
is different from the one in the summaries, we first use a fuzzy set-
based mapping operation to translate predicates of
Q
from the user-specific vocabulary to the summary
language. It consists in defining an accepting similarity threshold
τ
to decide whether the mapping of
two fuzzy labels is valid. In other words, the user's label
l
u
is rewritten with a summary label
l
s
if and
only if the similarity of
l
u
to
ls
(
σ(l
u
, l
s
)
) is greater than or equal to
τ
. There are many propositions in the
literature for defining
σ(l
u
, l
s
)
(e.g., degree of satisfiability (Bouchon-Meunier, Rifqi & Bothorel, 1996)).
Then, the
ESRA
Algorithm is performed using the rewritten version of
Q
. Finally, results are sorted and
flirted thanks to their similarity degrees to the initial user's query
Q
.
3.6 Experimental Results
Evaluating and comparing the effectiveness of different approaches that address the
Many-Answers
problem in databases is challenging. Unlike Information Retrieval where there exist extensive user
studies and available benchmarks (e.g., the TREC
xviii
collection), such infrastructures are not available